hidden_sizes: [128, 128]
total_timesteps: 5000000
rollout_steps: 1024
learning_epochs: 4
mini_batches: 4
discount_factor: 0.99
gae_lambda: 0.95
learning_rate: 0.0003
clip_ratio: 0.2
value_loss_scale: 0.5
entropy_loss_scale: 0.05
log_interval: 1000
checkpoint_interval: 50000

initial_log_std: 0.5
min_log_std: -2.0
max_log_std: 2.0

record_video_every: 10000

# ClearML remote execution (GPU worker)
remote: false

# ── HPO search ranges ────────────────────────────────────────────────
# Read by scripts/hpo.py — ignored by TrainerConfig during training.
hpo:
  learning_rate: {min: 0.00005, max: 0.001}
  clip_ratio: {min: 0.1, max: 0.3}
  discount_factor: {min: 0.98, max: 0.999}
  gae_lambda: {min: 0.9, max: 0.99}
  entropy_loss_scale: {min: 0.0001, max: 0.1}
  value_loss_scale: {min: 0.1, max: 1.0}
  learning_epochs: {min: 2, max: 8, type: int}
  mini_batches: {values: [2, 4, 8, 16]}